Remember Harry? Your golf buddy's son who started at the front desk and ended up approving invoices? (Missed it? Catch up here: Your AI Has More Authority Than You Approved.) He was a good illustration of how AI (or a receptionist) can drift into more authority than you intended.
In this article, we will discuss what to do before your rollout goes any further. You can do this by asking four questions. That's it. If your leadership team can answer all four today, you're in rare company. A survey by CSA and Google Cloud found only 26% of organizations had comprehensive AI security governance policies in place. If you can't answer them, don’t stress; I am here to help!
Let's dive in, shall we?
Who’s Driving? (The Series)
This is the second article in our AI series. In this series, we focus on AI adoption because nearly every organization is rolling out AI right now, and most have no formal way to track how much authority it has. We lay out a straightforward framework to help you decide how much authority your AI gets, spot the risks, and guard against them before something goes sideways.
Question 1: What AI Is Running Here That We Didn't Approve?
I am not referring to the tools on your approved list. I'm talking about the browser extension someone installed last Tuesday, the AI note-taker your CEO uses on every video call, the feature your vendor switched on over the weekend, and the free AI account on Harry's phone.
Pro-Tip
Ask your people what AI tools they use to get work done. Keep it a conversation, not an interrogation, and they’ll tell you things no scan will.
How big is this? Harmonic Security, which tracks how employees use AI tools, found the average enterprise saw 23 previously unknown AI tools newly used by employees in a single quarter.
You can't govern what you don't know about.
Bottom line: Your people are likely using more AI tools than you approved. Find the unapproved ones, then decide whether to approve, replace, or block each.
Question 2: Can Sensitive Data Go Into These Tools, and Has Anyone Written That Down?
By sensitive data, I mean: customer records, employee files, contract terms, login credentials, etc. Somebody in your organization is deciding right now whether to paste that into an AI tool. Actually, a lot of somebodies.
In a recent Zapier survey of 1,005 U.S. workers at companies with 500+ employees, they determined that:
89% surveyed had entered company data into AI
79% had entered especially sensitive information, like HR data, logins, or customer and employee personal information
Here's the kicker. In the same survey, 85% said they usually consider whether information is safe before entering it. So people ARE thinking about it. They're just answering the question on their own, one paste at a time, with no guidance from you.
If you have 250 employees and no written answer, you don't have a policy. You have 250 policies with zero governance.
So what should that written answer cover? It doesn’t need to be a 40-page policy. A one-page guide that tells employees what they can and can’t put into AI tools will do the job, as long as it’s easy to find and covers four things:
What can’t go in. Name the categories, like customer and employee personal information, financial records, login credentials, and contract terms. A good rule of thumb is: if you’d redact it before sharing it outside the company, redact it before pasting it into AI.
Which tools and accounts are approved. Tell people to use the company account, never a personal one. A personal account sits outside your company’s contract and protections, so you can’t help when something goes wrong.
What to do in the meantime. If your answer is still “not yet,” tell people what they can use AI for today, like drafting a generic email or summarizing a public document.
Who to ask. Name the person your people should go to when they’re unsure whether something is safe to paste.
Bottom line: Decide what to do with sensitive data before you do a full AI rollout. Once you decide, write it down and share it so everyone knows the policy.
Question 3: What Can Each Tool Reach Today, and Who Approved That?
Remember the connectors from the last article? Every AI tool can see something: your inbox, your file shares, your CRM, your calendar. If a tool can see it, someone allowed that. Or nobody within your organization did, and the vendor just turned it on.
The answer you'll hear most often? "It came that way." That’s not an explanation. It tells you nobody at your company decided what that tool could reach, and now you need to make that decision.
Pro-Tip
List what each AI tool can access and who approved that access. Any tool with no approver goes at the top of your list to address.
Let’s assume you own a 250-person property management firm. You approved an AI assistant to draft tenant emails. Can it also read the shared drive with every lease, every tenant's income verification, and every vendor's bank details? If nobody can tell you, assume yes.
Bottom line: Determine the reach of your AI tools and assess if that level of access was approved and by whom.
Question 4: When Something Goes Wrong, How Would We Find Out?
I am not referring to what the policy says (this is important, too), but to the actual alert sent to someone.
Pro-Tip
Pick one AI tool and ask your team: if it did something it shouldn’t, who would know? If you don’t like the answer, fix it.
Say that tenant-email assistant reads a maintenance request with hidden instructions buried in it (prompt injection, which gets its own article in this series) and starts forwarding tenant records to an outside address. Who gets the alert? How fast? What about at 2 AM on a Saturday?
If the honest answer is "we'd find out when a tenant complains," well... that's an answer. Just not the one that keeps you out of hot water.
Bottom line: For every AI tool, know who gets alerted, and how fast, when the tool does something it shouldn’t.
In Conclusion
Four questions:
What AI is running here that we didn't approve?
Can sensitive data go into these tools, and has anyone written that down?
What can each tool reach today, and who approved that?
When something goes wrong, how would we find out?
None of them needs a new tool or a big budget. They need someone to ask, and someone to own the answer. Get all four answered before your rollout goes any further, and the rest of this series gets a whole lot easier.
If your leadership team can't answer all four today, that's exactly the conversation I have with organizations rolling out AI. Contact us; we can help!
To infinity and beyond… share this article with everyone once you get to infinity or beyond. LOL
